Condition Monitoring and Fault Diagnosis

Introduction

Condition monitoring is the systematic measurement and interpretation of machine condition so that developing faults can be detected before they cause unacceptable loss of function. Fault diagnosis goes further: it identifies the likely failure mechanism, its location, severity, and the action required. Together, these activities connect maintenance engineering with rotating machinery, materials, instrumentation, data analysis, and risk management.

A condition-monitoring programme is not simply a collection of sensors. It is a decision system comprising a defined asset, measurable failure modes, suitable technologies, reliable data, competent interpretation, and an agreed response. Its purpose may be early warning, planned repair, confirmation of repair quality, protection against catastrophic failure, or improvement of reliability models.

Learning Objectives

After studying this chapter, you should be able to:

1. Explain the principles and applications of major condition-monitoring technologies.
2. Select suitable monitoring methods for common mechanical and electrical failure modes.
3. Establish baselines, alarm limits, measurement routes, and diagnostic workflows.
4. Interpret vibration, oil, thermal, ultrasonic, acoustic-emission, motor-current, and nondestructive-testing results.
5. Distinguish symptoms from root causes and make defensible maintenance decisions.
6. Recognize measurement uncertainty, false alarms, safety obligations, and ethical responsibilities.

1. Condition Monitoring as a Reliability Process

Traditional preventive maintenance replaces or services equipment at fixed intervals. Condition-based maintenance instead schedules intervention when evidence indicates that degradation has reached a planned decision threshold. The approach is especially valuable where failure is gradual, monitoring is technically feasible, and the consequences of failure justify the cost.

A useful diagnostic chain is:

failure mode → physical symptom → measurement → feature → diagnosis → decision → verification

For example, a bearing race defect may produce impulsive impacts, which appear as high-frequency vibration and envelope-spectrum peaks at characteristic defect frequencies. The measurement is not itself the diagnosis; it is evidence that must be interpreted with speed, load, machine history, and other observations.

Figure: A condition-monitoring workflow from failure mode to verified maintenance decision.

Important terminology

• Baseline: a documented reference condition measured under defined operating conditions.
• Feature: a calculated or observed indicator, such as RMS velocity, oil particle count, or temperature difference.
• Alarm limit: a threshold requiring review or action.
• Trend: the change in a feature over time.
• Fault signature: a characteristic pattern associated with a failure mechanism.
• Diagnostic confidence: the strength and consistency of evidence supporting a diagnosis.

2. Establishing Baselines and Alarm Limits

A baseline should be collected from an asset known to be correctly installed, aligned, lubricated, and operating normally. Measurements must record operating state: rotational speed, load, temperature, product flow, ambient conditions, sensor position, instrument settings, and date. Comparing a lightly loaded machine with a fully loaded baseline can create a false alarm.

Baseline values are often represented by a central value and variation. For a feature measured as x₁, x₂, …, xₙ, the sample mean is:

Here x̄ is the mean, xᵢ is observation i, and n is the number of observations. The sample standard deviation is:

The divisor n − 1 is used when estimating population variability from a sample. These statistics are meaningful only when observations represent comparable operating conditions and are not contaminated by known abnormal events.

A statistical screening limit may be set as x̄ + k s for an upward-only feature, where k is selected according to risk and the desired false-alarm rate. This is not a universal engineering limit: machinery standards, manufacturer limits, criticality, consequences, and measurement uncertainty must also be considered. For non-normal data, robust percentiles or median-based methods may be preferable.

A practical alarm structure is:

Alarm limits should be validated against historical failures and reviewed after modifications. A limit must be specific about feature, units, sensor location, operating condition, time window, and action. “High vibration” is not an adequate alarm definition.

3. Vibration Analysis

Vibration is the most widely used technique for rotating machinery because faults alter dynamic forces and structural response. A transducer measures acceleration, velocity, or displacement. Acceleration is especially useful at high frequencies; velocity is widely used for general machine severity; displacement can be useful at low-frequency shaft motion.

For sinusoidal motion, displacement amplitude X, angular frequency ω, velocity amplitude V, and acceleration amplitude A are related by:

where ω = 2πf in rad/s and f is frequency in hertz. The relations apply to a single sinusoidal component; real machines contain multiple components. Unit conversion is essential: millimetres, metres, seconds, and hertz must be used consistently.

The root-mean-square value of a sampled signal xᵢ is:

For a zero-mean signal, RMS indicates energy-like magnitude. It does not identify the fault by itself. Crest factor, defined as peak divided by RMS, can help identify impulsive events, although it is sensitive to noise and sampling.

Frequency-domain interpretation

A time waveform shows when events occur; a spectrum shows their frequency content. Common components include:

• running speed, often called 1× rotational frequency;
• harmonics caused by misalignment, looseness, or waveform distortion;
• gear-mesh frequency and sidebands;
• bearing defect frequencies;
• electrical frequencies and twice-line-frequency components;
• resonance-related peaks.

Frequency alone is insufficient. Phase, direction, coherence, speed, load, and location improve diagnosis. A 1× peak may result from unbalance, bent shaft, eccentricity, resonance, or misalignment; the phase pattern and machine geometry distinguish these possibilities.

Envelope or demodulation analysis filters high-frequency bearing impacts and analyses their repetition rate. It is useful for rolling-element bearing defects, but poor mounting, inadequate bandwidth, lubrication noise, and variable speed can reduce reliability.

Worked example: unit consistency

A vibration component has displacement amplitude 12 μm at 50 Hz. Since 12 μm = 12 × 10⁻⁶ m and ω = 2π(50) = 314.16 rad/s,

V = ωX = 314.16 × 12 × 10^-6 = 0.00377 m/s

Thus the velocity amplitude is approximately 3.77 mm/s. This conversion applies to the stated sinusoidal component, not automatically to a broadband RMS reading.

4. Oil and Wear-Debris Analysis

Lubricant analysis evaluates both lubricant health and machine-generated contamination. Samples should be taken from a live, well-mixed zone using clean containers and a repeatable method. A drain-port sample taken after sediment has settled may not represent circulating oil.

Typical tests include viscosity, water content, oxidation, acid number, additive depletion, particle count, elemental spectroscopy, ferrography, and analytical ferrography. Results can indicate:

• dirt ingress or inadequate filtration;
• water contamination and corrosion risk;
• abnormal ferrous or non-ferrous wear;
• lubricant mixing or wrong-grade oil;
• overheating, oxidation, or additive exhaustion.

Particle-count codes such as ISO cleanliness codes describe size ranges and must not be confused with a direct measure of wear severity. Spectroscopy detects very small particles effectively but may underrepresent large particles. Ferrography can reveal morphology and severity but requires skilled interpretation.

The most useful result is usually a trend tied to known oil changes, filter replacements, operating events, and component identity. A high value without sample provenance is weak evidence.

5. Infrared Thermography

Thermography converts infrared radiation into an apparent surface-temperature image. It is non-contact and valuable for electrical connections, bearings, couplings, lubrication problems, insulation defects, steam systems, and heat exchangers.

Temperature interpretation depends on emissivity, reflected radiation, atmospheric path, angle, focus, and load. Shiny metal may produce large errors because its emissivity is low and it reflects surrounding objects. Whenever possible, compare similar phases, components, or machines under similar load rather than relying on an absolute temperature alone.

A useful comparative quantity is temperature difference:

where T_hot is the measured temperature of the suspect location and T_reference is a comparable sound location. The comparison is meaningful only when locations have similar construction, load, airflow, and emissivity. Thermography identifies a thermal symptom; it does not by itself prove whether the cause is resistance, overload, poor cooling, friction, or external heating.

6. Ultrasonics and Acoustic Emission

Ultrasonic instruments detect high-frequency sound, commonly associated with turbulent flow, compressed-air leakage, electrical discharge, and frictional contact. They are useful in noisy industrial environments because ultrasound can be filtered and converted to an audible signal. Leak surveys should be performed with a defined distance, gain, background condition, and load.

Acoustic emission (AE) differs from routine ultrasonic inspection. AE detects transient elastic waves released by sudden energy changes, such as crack growth, friction, impact, or fibre breakage. Sensors, coupling, threshold, arrival time, and multiple-channel location affect results. AE is therefore a dynamic event-detection method rather than simply a temperature or steady-noise measurement.

Ultrasonic and AE results are strongly affected by attenuation, reflections, sensor coupling, and background sources. Confirmation by visual inspection, vibration, process data, or nondestructive testing is often necessary.

7. Motor-Current Signature Analysis

Motor-current analysis measures current and voltage to infer electrical and mechanical condition without direct access to the machine. It can identify supply imbalance, broken rotor bars, eccentricity, misalignment, overload, and some driven-load problems.

The technique is sensitive to motor design, slip, load, power quality, inverter operation, and frequency resolution. A spectral sideband should not be interpreted in isolation. Variable-frequency drives introduce switching components and changing fundamental frequency, requiring appropriate acquisition and specialist analysis. Current analysis is particularly attractive where vibration sensors are difficult to install, but it is usually a complementary rather than universal technique.

8. Nondestructive Testing

Nondestructive testing (NDT) examines material or component integrity without destroying its serviceability. Selection depends on the suspected discontinuity, material, geometry, accessibility, required sensitivity, and safety implications.

NDT should be selected from a failure hypothesis. For example, dye penetrant is appropriate for a suspected surface crack in a non-porous weld, whereas ultrasonic testing may be more suitable for an internal lack of fusion. Inspection personnel must be qualified under the governing employer, national, or sector certification system.

9. Diagnostic Workflow and Data Quality

A disciplined workflow is:

1. Define the asset, function, criticality, and credible failure modes.
2. Select measurable symptoms and suitable technologies.
3. Establish a baseline under documented operating conditions.
4. Collect repeatable measurements using calibrated instruments.
5. Validate abnormal results and exclude changed operating conditions.
6. Compare trends, spectra, images, laboratory results, and inspection findings.
7. Form and rank hypotheses by evidence and consequence.
8. Decide whether to continue, increase monitoring, plan repair, or protect/shut down.
9. Verify the diagnosis after intervention.
10. Record the result and update the failure model and alarm settings.

Data quality has technical and organisational dimensions. Sensor orientation, mounting torque, cable condition, calibration, sampling rate, anti-alias filtering, time synchronisation, and metadata all matter. A sophisticated algorithm cannot recover information that was never measured.

When several indicators disagree, do not average them blindly. Investigate sensor error, operating changes, cross-sensitivity, and different degradation stages. A vibration alert with normal oil results may indicate a mechanical dynamic problem, a sampling issue, or a fault too early to generate wear debris.

10. Integrated Diagnosis and Decision-Making

Each technology has a characteristic evidence domain:

Diagnosis should be expressed with confidence and uncertainty: “probable outer-race bearing defect, supported by repeated envelope peaks and increasing temperature; confirm during planned inspection.” Such wording is more professional than claiming certainty from a single measurement.

Decision priority should combine condition, consequence, detectability, and time to functional failure. A modest change on a safety-critical compressor may deserve faster action than a larger change on a redundant, low-consequence pump. The monitoring programme must therefore be integrated with the asset criticality analysis and maintenance strategy.

11. Illustrative Case: Conveyor Drive Bearing

Background. A mineral-processing conveyor uses a 90 kW motor, flexible coupling, gearbox, and driven pulley. The drive-end bearing is monitored monthly by vibration and quarterly by oil analysis.

Problem. Overall velocity increases from 2.1 to 4.0 mm/s RMS over three months. The machine speed and throughput are unchanged. A thermographic image shows the bearing housing 18 °C hotter than the comparable non-drive-end housing.

Analysis. A repeat measurement confirms the vibration increase. The spectrum contains a growing high-frequency component and envelope peaks consistent with the bearing geometry. Oil analysis shows a moderate increase in ferrous debris but no water contamination. The coupling and base are inspected; looseness is not found. The evidence supports progressive bearing damage rather than a simple alignment symptom.

Approach. The planner checks the remaining production window and orders a bearing, seal, and lubricant. Monitoring frequency is increased to weekly. A controlled shutdown is scheduled before the next high-load campaign. During replacement, the bearing shows spalling on the outer race and evidence of inadequate lubrication. The housing fit and shaft are inspected before assembly, and installation tools are used to avoid transmitting force through the rolling elements.

Outcome. After commissioning, vibration returns to 1.9 mm/s RMS and the temperature difference falls to 4 °C. The lubrication route is revised, sample points are improved, and the alarm response requires confirmation of lubrication and load before escalation.

Learning points. The decision was based on a trend and multiple evidence sources, not one threshold. The baseline included comparable operating conditions. The final inspection verified the diagnosis and converted the event into a reliability improvement.

12. Limitations, Governance, and Contemporary Practice

Condition monitoring cannot detect every failure. Some failures are sudden, hidden, intermittent, or not physically observable by available sensors. Measurement systems can generate false positives and false negatives. Alarm fatigue occurs when thresholds are too sensitive or poorly maintained. Conversely, excessive reliance on historical baseline values can miss a consistently degraded machine.

Safety is primary. Inspectors must respect rotating equipment, electrical arc-flash hazards, radiation controls, hot surfaces, chemical exposure, and confined-space rules. A monitoring activity must never require unsafe access merely to obtain a preferred measurement.

Records should preserve raw data, processed features, instrument identity, calibration status, analyst, operating state, alarm changes, decisions, and verification results. Access controls and audit trails are important when data support regulatory compliance, warranty claims, or safety decisions. Automated and AI-based anomaly detection can prioritize large data streams, but models require representative training data, explainable features, drift monitoring, and human review. An anomaly is not automatically a diagnosis.

Contemporary systems increasingly combine wireless sensors, edge processing, digital asset models, and machine-learning methods. These technologies are valuable when they improve coverage and response, but they do not remove the need for failure-mode analysis, physical understanding, calibration, cybersecurity, and competent engineering judgement.

Best-Practice Checklist

• Start with failure modes and consequences, not with available instruments.
• Standardise sensor locations, operating conditions, units, and metadata.
• Establish a sound baseline and document its validity.
• Use trends and corroborating evidence rather than isolated values.
• Set alarm limits linked to defined actions and review them periodically.
• Calibrate instruments and verify abnormal readings.
• Use qualified personnel for specialist interpretation and NDT.
• Integrate findings with work management, spares, risk, and production planning.
• Verify the diagnosis after repair and feed lessons back into the programme.

Chapter Summary

Condition monitoring transforms physical symptoms into maintenance decisions through measurement, interpretation, and verification. Vibration analysis is central to rotating-equipment diagnosis, while oil analysis reveals lubricant and wear condition; thermography reveals heat patterns; ultrasonics and acoustic emission detect high-frequency leakage, friction, and transient damage; motor-current analysis provides electrical and indirect mechanical evidence; and NDT examines material integrity.

Reliable programmes require valid baselines, repeatable measurements, context-sensitive alarm limits, competent diagnosis, and documented responses. No technology is universally sufficient. The strongest decisions integrate trends, operating conditions, multiple evidence sources, criticality, and post-maintenance verification. Digital and AI technologies can expand coverage, but engineering judgement, safety governance, and physical understanding remain essential.

Key Terms

Baseline: Reference condition measured on a healthy asset under defined conditions.

Condition monitoring: Measurement and interpretation of asset condition over time.

Fault diagnosis: Identification of the likely mechanism, location, and severity of a fault.

Feature: A measured or calculated indicator extracted from data.

Alarm limit: A defined threshold that initiates review or action.

Trend: Direction and rate of change of a condition indicator.

Vibration RMS: Root-mean-square magnitude of a vibration signal.

Crest factor: Peak amplitude divided by RMS amplitude.

Envelope analysis: Demodulation method used to reveal repetitive impacts.

Wear debris: Particles generated by material removal during operation.

Ferrography: Analysis of ferrous particles separated from lubricant.

Emissivity: Effectiveness of a surface in emitting infrared radiation.

Ultrasonics: Use of high-frequency sound for condition assessment.

Acoustic emission: Transient elastic waves released by sudden material or frictional events.

Motor-current signature analysis: Interpretation of motor-current features to assess electrical and mechanical condition.

Nondestructive testing: Examination of materials without impairing future serviceability.

False positive: Alarm indicating a fault when the asset is not faulty.

False negative: Failure to alarm when a significant fault exists.

Criticality: Relative consequence of asset failure.

Diagnostic confidence: Strength of evidence supporting a diagnosis.

Review Questions

Short Answer Questions

1. What is the difference between condition monitoring and fault diagnosis?
2. Why must a baseline include operating conditions?
3. State two common applications of vibration analysis.
4. Why can particle-count data not be treated as a complete measure of wear severity?
5. Name two factors that affect infrared thermography accuracy.

Long Answer Questions

1. Explain how an integrated condition-monitoring programme is designed from failure modes to verified maintenance action.
2. Compare vibration analysis, oil analysis, thermography, ultrasonics, and motor-current analysis in terms of evidence produced and limitations.
3. Describe how alarm limits should be established, validated, and revised.
4. Explain the role of sampling quality, calibration, metadata, and operating context in condition-monitoring reliability.
5. Discuss the benefits and risks of applying AI or machine learning to condition-monitoring data.

Analytical / Application Questions

1. A pump shows a growing 1× vibration component, but oil analysis and temperature remain normal. Propose at least three hypotheses and identify additional measurements needed to discriminate between them.
2. A gearbox oil sample contains increased ferrous debris, while vibration remains unchanged. Explain possible reasons for this apparent disagreement and recommend a diagnostic plan.
3. Design a monitoring workflow for a safety-critical fan driven by a variable-frequency motor. Specify measurements, baseline conditions, alarm responses, and verification steps.

Multiple Choice Questions with Answers

Question 1

Which statement best describes a condition-monitoring baseline?

A. The highest value ever recorded on the asset

B. A reference condition measured under defined, normal operating conditions

C. A universal alarm value supplied by any instrument manufacturer

D. The average value from unrelated machines

Correct Answer: B. A reference condition measured under defined, normal operating conditions

Question 2

For a sinusoidal vibration component, which relationship is correct?

A. V = X/ω

B. V = ωX

C. V = ω²X

D. V = X²/ω

Correct Answer: B. V = ωX

Question 3

Which method is particularly useful for detecting lubricant contamination and wear particles?

A. Oil and wear-debris analysis

B. Thermography only

C. Visual inspection of paint colour

D. Motor insulation resistance only

Correct Answer: A. Oil and wear-debris analysis

Question 4

A shiny metal surface produces an unreliable infrared temperature reading mainly because of its:

A. High electrical resistance

B. Low rotational speed

C. Low emissivity and reflected radiation

D. High oil viscosity

Correct Answer: C. Low emissivity and reflected radiation

Question 5

Which NDT method is generally suitable for surface-breaking cracks in a non-porous, non-magnetic component?

A. Liquid penetrant testing

B. Magnetic-particle testing

C. Radiographic testing only

D. Motor-current analysis

Correct Answer: A. Liquid penetrant testing

Question 6

What does envelope analysis commonly help reveal in rolling-element bearings?

A. Repetitive impact frequencies associated with defects

B. Lubricant viscosity directly

C. Surface emissivity

D. Electrical supply voltage

Correct Answer: A. Repetitive impact frequencies associated with defects

Question 7

A measurement trend rises after the machine load changes substantially. The first diagnostic action should be to:

A. Ignore all future measurements

B. Compare the result with the baseline and account for the changed operating condition

C. Replace every bearing immediately

D. Delete the earlier baseline

Correct Answer: B. Compare the result with the baseline and account for the changed operating condition

Question 8

Which statement about motor-current signature analysis is most accurate?

A. It is unaffected by motor load or variable-frequency drives

B. It can provide evidence of electrical and some mechanical problems but requires contextual interpretation

C. It directly measures bearing surface cracks in every machine

D. It replaces all forms of vibration measurement

Correct Answer: B. It can provide evidence of electrical and some mechanical problems but requires contextual interpretation

Question 9

A displacement amplitude is 12 μm at 50 Hz. What is the approximate velocity amplitude?

A. 0.19 mm/s

B. 1.20 mm/s

C. 3.77 mm/s

D. 37.7 mm/s

Correct Answer: C. 3.77 mm/s

Question 10

Why should an abnormal condition-monitoring result normally be verified before major intervention?

A. Because all alarms are necessarily incorrect

B. To distinguish a genuine fault from measurement error or changed operating conditions

C. To prevent any maintenance planning

D. Because trends have no diagnostic value

Correct Answer: B. To distinguish a genuine fault from measurement error or changed operating conditions